Build the factories and the boom will follow: that is OpenAI’s bet as it inks a $38 billion agreement to run its systems on Amazon Web Services, part of a mammoth plan to secure more than $1 trillion of computing capacity.
The deal gives OpenAI immediate access to AWS datacentres and the Nvidia processors inside them, the specialist chips that train and run models such as ChatGPT.
Amazon.com Inc (NASDAQ:AMZN) said it will cluster “hundreds of thousands” of these graphics processors for OpenAI, powering day-to-day responses and the next generation of models.
In parallel, OpenAI is committing to develop about 30 gigawatts of computing resources, a grid-scale push roughly equivalent to the electricity needs of 25 million US homes.
The spend dwarfs the company’s reported revenue run-rate and has revived a simple question: who pays? OpenAI has just reorganised into a for-profit corporation at a $500 billion valuation, with Microsoft poised to own about 27%.
It has also signed separate infrastructure arrangements, including a $300 billion agreement with Oracle.
Chief executive Sam Altman has brushed off worries about the gap between income and outlays, arguing that scaling “frontier” AI requires reliable capacity first and that investor appetite remains deep.
For readers keeping score, the numbers are moving fast. Adjusted for timing and partners, OpenAI’s commitments sit within a broader data-centre arms race: Wall Street estimates put global spend near $3 trillion by 2028, with Big Tech funding roughly half and private credit increasingly stepping in for the rest.
The prize is straightforward, faster, more capable AI, but so are the constraints: power, chips and capital, in that order.